EAGER: Fairness-Aware Personalized Recommendations
NSF IIS: 1841138, August 2018 to July 2020
Project Goals: The goal of this project is to create effective recommendation models that can be personalized for individual users, while maintaining important fairness properties. A significant emphasis is on information curators and their items. The advances in uncovering information curators at scale, reliably connecting users to appropriate curators, and ensuring fairness-preserving properties of such curators are critical for trustworthy information supporting an informed populace. By bringing these research advances, datasets, and toolkits to the wider research community, this project can spur additional advances from complementary efforts by other researchers.
Participants:
Research Challenges: A key challenge for personalized recommendation is tackling sparsity while carefully modeling curators, items, and users in complex, noisy, and heterogeneous environments. Compounding this challenge, most current access to information curators is mediated by centralized platforms (like search engines, social networks, and traditional news media), meaning that personal preferences may not align with the goals of these platforms, leading to potentially biased (or even limited) access to curators. A key question is how to maintain fairness properties in personalized recommendation.
Broader Impacts: The successful outcome of this project will lead to research advances that can positively impact existing web and social media platforms, as well as provide a theoretical foundation for future advances in recommendation. The advances in uncovering information curators at scale, reliably connecting users to appropriate curators, and ensuring fairness-preserving properties are critical for a trustworthy information diet supporting an informed populace. This project will develop new classroom materials, new outreach efforts, and new broadening participation workshops and seminars. Together, these efforts will integrate the new knowledge developed as part of the research plan through investments in undergraduate and graduate students, and through course enhancements and research training.
Current Results: In the two years of this project, our research team made a number of strides in improving recommendation, guided by our proposed research effort. Concretely, we have focused on: 1) new neural models of personalized recommendation; 2) new fairness-aware approaches for recommendation; and 3) new recommendation applications.
- In our first research thrust, our goal is to create new neural models for personalized recommendation. We have created (i) new disentangled representation learning models for improved recommendation, (ii) a randomized training and mixture-of-experts approach for the cold start problem, (iii) a personalized user recommendation framework for content curation platforms that models preferences for both users and the items they engage with simultaneously, (iv) a novel recommendation system that explicitly captures the influence from key opinion leaders to the whole community, (v) a new time-dependent neural predictive model for personalized recommendation that balances the long-term evolution of a user's interests with short-term "local coherence", and (vi) a new Joint Collaborative Autoencoder framework that learns both user-user and item-item correlations simultaneously, leading to a more robust model and improved top-K recommendation performance. This effort has resulted in papers appearing at RecSys 2020, SIGIR 2020, WSDM 2020 (x2), WWW 2019, and WSDM 2019.
- In our second research thrust, our goal is to augment personalized recommendation under fairness-aware constraints. Since recommenders may inherit bias from the training data used to optimize them and from mis-alignment between platform goals and personal preferences, we have been exploring new fairness-aware algorithms that can empower users. Specifically, we have proposed (i) how to generate unbiased recommendations based on biased implicit user-item interactions through a combinational joint learning framework to simultaneously learn unbiased user-item relevance and unbiased propensity, (ii) a debiased personalized ranking model that is based on a ranking-based statistical parity and equal opportunity as two measures of item under-recommendation bias, and (iii) a new method to augment tensor-based recommenders with statistical parity-based fairness constraints. This effort has resulted in papers appearing at RecSys 2020, SIGIR 2020, CIKM 2018, and the 2nd FATREC Workshop on Responsible Recommendation at RecSys, 2018.
- Finally, we have been investigating new application areas in personalized recommendation. In particular, we have developed (i) a novel next-item recommendation framework
empowered by sequential hypergraphs, (ii) a new approach for recommending user-generated item lists by intelligently combining two preference models via a novel consistency-aware gating network – a general user preference model that captures a user's overall interests, and a current preference priority model that captures a user’s current (as of the most recent item) interests, (iii) another effort focusing on item lists that focuses on item and list consistency, through a novel self-attentive aggregation layer designed for capturing the consistency of neighboring items and lists to better model user preference, and (iv) the first visual influence-aware fashion recommender that leverages fashion leaders and their dynamic visual posts. This effort has resulted in papers appearing at SIGIR 2020, WSDM 2020, and CIKM 2019 (x2).
Publications:
- Yin Zhang, Ziwei Zhu, Yun He, and James Caverlee. Content-Collaborative Disentanglement Representation Learning for Enhanced Recommendation. RecSys 2020.
- Ziwei Zhu, Yun He, Yin Zhang, and James Caverlee. Unbiased Implicit Recommendation and Propensity Estimation via Combinational Joint Learning (short paper). RecSys 2020.
- Ziwei Zhu, Jianling Wang, and James Caverlee. Measuring and Mitigating Item Under-Recommendation Bias in Personalized Ranking Systems. SIGIR 2020
- Ziwei Zhu, Shahin Sefati, Parsa Saadatpanah, and James Caverlee. Recommendation for New Users and New Items via Randomized Training and Mixture-of-Experts Transformation. SIGIR 2020.
- Jianling Wang, Kaize Ding, Liangjie Hong, Huan Liu, and James Caverlee. Next-item Recommendation with Sequential Hypergraphs. SIGIR 2020.
- Yun He, Yin Zhang, Weiwen Liu, and James Caverlee. Consistency-Aware Recommendation for User-Generated Item List Continuation. WSDM 2020.
- Jianling Wang, Ziwei Zhu, and James Caverlee. User Recommendation in Content Curation Platforms. WSDM 2020.
- Jianling Wang, Kaize Ding, Ziwei Zhu, Yin Zhang, and James Caverlee. Key Opinion Leaders in Recommendation Systems: Opinion Elicitation and Diffusion. WSDM 2020.
- Yun He, Jianling Wang, Wei Niu, and James Caverlee. A Hierarchical Self-Attentive Model for Recommending User-Generated Item Lists. CIKM 2019.
- Yin Zhang and James Caverlee. Instagrammers, Fashionistas, and Me: Recurrent Fashion Recommendation with Implicit Visual Influence. CIKM 2019.
- Ziwei Zhu, Jianling Wang, and James Caverlee. Improving Top-K Recommendation via Joint Collaborative Autoencoders (short paper) WWW 2019.
- Jianling Wang and James Caverlee. Recurrent Recommendation with Local Coherence, WSDM 2019.
- Ziwei Zhu, Xia Hu, and James Caverlee. Fairness-Aware Tensor-Based Recommendation, CIKM 2018.
- Ziwei Zhu, Jianling Wang, Yin Zhang and James Caverlee. Fairness-Aware Recommendation of Information Curators, The 2nd FATREC Workshop on Responsible Recommendation at RecSys, 2018.
This material is based upon work supported by the National Science Foundation under Grant No. 1841138. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.
Date of Last Update: August 2020